Last week, New York City hosted two events which wear out the shoe leather (and sometimes the patience) of New Yorkers and visitors alike: the United Nations General Assembly (UNGA), and Climate Week NYC.
In April, I attended an SF Climate Week panel on the ways in which artificial intelligence solutions could support the transition to net zero electricity. When the session opened for comments, I was a bit surprised that the first ten audience members confronted the four panelists with variations of the exact same pointed question: “Aren’t your data centers going to overwhelm the electric grid and ensure climate disaster?”
The 11th and final commentator came in equally hot, but in the opposite direction. He dropped all pretense of asking a question and addressed the crowd directly (paraphrasing): “You all don’t know what you’re talking about! Electric vehicles (EVs) will be ten times bigger than data centers, and we’re all going to be fine.”
The exchange highlighted a tension that I’ve noticed across the energy world: everyone is worried that data centers will use more electricity than we can deliver. They’re worried we won’t be able to meet data center power needs from renewable sources, and they are worried we’ll demand more electricity than the transmission grid can move. A quick Halcyon query confirms that some of energy’s most central regulators share these concerns:
As the 11th commenter pointed out, one rarely hears these same concerns about the rise of electric vehicles. But are those concerns warranted? A bit of quantitative analysis can help us answer this question.
Evaluating relative electricity use between the two today is straightforward. The Federal Energy Regulatory Commission (FERC) estimates that in the U.S., data centers consumed 126 terawatt-hours (TWh) of electricity in 2023. Goldman Sachs 2023 estimates were a bit higher at 146 TWh, but of that, only 4 TWh — or 2.7% of total data center electricity usage — was consumed specifically for AI. A lot of smart people reasonably expect that sub-sector to grow significantly.
Meanwhile, the EIA tells us that U.S. EVs used 7.6 TWh of electricity in 2023 — a mere 5.2% of data center consumption. So, today, data centers use about 20x more electricity than EVs.
Halcyon can bring authoritative data (with citations!) to quickly verify this analysis:
However, as the response hints, EV demand is expected to grow faster than data center electricity use over time. Napkin math time: there were ~3.3M EVs on U.S. roads in 2023. If we break down electricity consumption on a per-car basis, a U.S. EV uses about 2,300 kWh of electricity annually.
In a high-end scenario, let’s imagine that all US cars are replaced with an equivalent electric vehicle. There are around 280 million registered automobiles in the US, so assuming the same annual electricity consumption per car, replacing them all will net out to 644 TWh of consumption per year, which is about 5 times more than today’s data centers and 1.6 times greater than the 397 TWh that Goldman Sachs projects data centers will use by 2030.
However, there are reasons that EVs may never grow that much. Other fuels such as hydrogen, biofuel, or old fashioned petroleum may power some cars and trucks for decades into the future. Furthermore, the rise of rideshare and autonomous driving make it very hard to predict how many cars the US will need in the future, and how many miles they will each drive per year.
The future, clearly, is uncertain. It seems most likely that SF Climate Week event’s 11th commenter is probably wrong that EVs would categorically outpace data center electricity demand, but there are certainly multiple futures where his assumption bears out. There are reasonable scenarios where EVs accelerate to rival or even exceed data centers as a source of electricity demand, but none in which they are “ten times” that of data centers. In most scenarios they will have comparable demand.
Keeping these future scenarios in mind can help utilities, clean energy developers, hyperscale tech companies and other players in energy build an adaptable strategy. In all cases, we need to build a lot of new clean generation and many thousands of miles of new bulk transmission capacity, and do it now.
As regular readers know, there is an informational element to building with speed and scale, and The information that determines how that infrastructure gets built, by whom, where, and when, is exactly what Halcyon is trying to make easier to access.
Comments or questions? We’d love to hear from you - sayhi@halcyon.eco, or find us on LinkedIn and Twitter